Peter Daley, managing director at rFpro, asks how far the automotive industry can go down the simulation route
The global pandemic has accelerated the trend for organizations to re-examine the split between physical testing and virtual simulation. How do you maintain progress in development and validation when key staff are working from home or access to test facilities is limited?
The decision between physical and virtual testing is based on relative cost, speed and reliability, but, as technologies advance, the preferred option may change. In many fields, simulations that were too slow or unreliable a few years ago are now becoming established as part of normal product development.
The importance of driving simulation in the development and testing of vehicles has grown massively in recent years and we can expect it to continue growing for some time to come. It can be done earlier in the design process, more quickly, at lower cost, more safely and in greater quantity and variety than real-world testing. Techniques adopted initially by professional motorsport teams, with the aim of maximizing speed around a track, are now increasingly being adapted and used by the automotive industry to develop and optimize a wide range of attributes in the latest generation of road vehicles.
Correlation is key
The prerequisites for successful simulation include reliable data, often correlated with earlier physical tests or validated through subsequent prototype testing, and a clear understanding of the purpose and the scope. Preparation is crucial, including the creation of appropriate models and test scenarios, as well as planning for calibration against real-world tests and consequent model adjustments.
A current challenge for the industry is how to make driving simulation effective and efficient in the development and testing of autonomous vehicles. The growing trend to ADAS and AVs has massively increased the software development workload for auto makers and neatly illustrates the requirement for both virtual and physical testing. Virtual testing is an absolute necessity given the amount and nature of testing required to provide assurance that AVs are safe for use in a public environment.
Because AI is self-learning, it must be exposed to the random events that occur on public roads to accumulate the necessary experience, but this is too risky with an immature system. Effective simulation is the safest and quickest way to accelerate the number of scenarios experienced by an AV during early development. Physical track testing contributes during development of the simulation models, then again in validating the results. Finally, public road evaluation provides the ultimate confidence that no test scenarios have been overlooked during development.
Simulation of AVs presents many challenges. Overcoming them will require collaboration across a range of organizations including vehicle manufacturers, component and system suppliers, service providers, technology providers, regulatory bodies and governments, not to mention ordinary road users.
For driver-in-the-loop systems, full-motion simulators with high-quality, wraparound image projection create a high level of driver immersion. However, this comes at a cost and the level of immersion needed can vary depending on the application. In its most basic form, a driving simulation can run on a single workstation with one screen and yet still provide much valuable information provided that the modeling is realistic and accurate. When testing ADAS and AV systems, defining a suitable range of test scenarios and ‘edge cases’ is often, in itself, a challenge, as is creating and executing such test scenarios and edge cases in a suitable range of virtual environments and at a sufficient scale.
Achieving the necessary realism
The virtual environments must incorporate high levels of detail and fidelity in terms of geometric, visual, lidar and radar characteristics, including accurate sensor models. The modelers need to define and build a suitable range of test scenarios, including the ability to explore a multitude of variations within each scenario. Traffic, people, animals, stray objects and anything else that occurs in the real world have to be included, and varied. The same goes for weather and light conditions.
To achieve these things in a way that is realistic, a high level of sophistication is necessary. In rFpro’s virtual world, vehicular and pedestrian traffic can share the road network correctly, following the rules of the road, and experience ad-hoc behavior – such as pedestrians stepping into the road – to provoke an emergency. This enables digital experiments to precisely mirror the real-world tests conducted in the physical environment. Human test drivers can even interact with the virtual world by testing cars with ADAS, riding as a passenger in an AV, or simply driving around the virtual world to provoke and evaluate responses from AVs.
Using a full-motion driving simulator has become sufficiently realistic for vehicle dynamics development and assessment, thanks to the faster responses of the latest generation of motion platforms and the greatly increased fidelity of road surface modeling, such as rFpro’s Terrain Server. Realistic modeling of road surfaces also makes it possible to use a simulator to evaluate human factors such as ride comfort, and man-machine interfaces such as operating a touchscreen on a bumpy road.
The future for driving simulation is bright. It will continue to be used in professional motorsport while being increasingly adopted across the automotive industry and its supply chain, both in the traditional area of vehicle dynamics and beyond. Potential new areas include NVH and HMI, with rFpro’s recent developments in headlight profiling simulation being a good example of the latter.
It is widely recognized that application of simulation to the development of ADAS and AV technology is an absolute requirement. The sophistication and power of those technologies and the simulation tools they use will evolve together. In simulation, the deployment of high-performance computing, the increasingly accurate rendering of light and color, and the ability to generate multiple variations of critical scenarios quickly and efficiently will all play a part.
Will virtual simulation ever completely replace physical testing? Almost certainly not. Firstly, virtual tests always incorporate some kind of mathematical model that aims to mimic the way that things behave in the real world. The model will inevitably involve some level of simplification and assumption, so has to be repeatedly checked for correlation with the real world and recalibrated as necessary. Secondly, the real world can be complex, chaotic and inconsistent. Not everything can be accurately and reliably captured in a mathematical model. Nevertheless, the more simulation we do, the more we learn. As the power of available IT tools continues to evolve and our models become ever more faithful to reality and applicable across an ever wider set of use cases, our ability to model the physical world in a virtual environment will continue to grow.